Conditional Estimations for Seamless Phase II/III Clinical Trials Involving Multi-Stage Early Stopping

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Inference

Background:

• Seamless phase II/III designs aim to reduce trial duration and costs.
• Incorporating multi-stage early stopping in phase III can prevent sample size overestimation.
• Existing estimation methods are limited to two-stage designs and do not support multi-stage early stopping in phase III.

Purpose of the Study:

• Develop a Score-statistics-based framework for point and interval estimation in seamless phase II/III designs with multi-stage early stopping.
• Provide design-specific conditional bias adjustment, median-unbiased estimation, and Rao-Blackwellization.
• Evaluate the performance of various conditional estimation procedures.

Main Methods:

• Developed a Score-statistics-based framework for estimation.
• Formulated conditional bias-adjusted estimators (CBAE-MI, CBAE-SI), conditional median unbiased estimators (CMUE-MLE, CMUE-ZERO), and the Rao-Blackwellized (RB) estimator.
• Derived confidence intervals based on CMUE-MLE, CMUE-ZERO, and RB.

Main Results:

• The proposed framework enables valid and efficient estimation of treatment effects conditional on trial continuation to phase III.
• Simulations evaluated estimator performance across different endpoint types and parameter configurations.
• The Rao-Blackwellized (RB) estimator demonstrated superior robustness and conservative coverage probability.

Conclusions:

• The developed framework effectively addresses estimation challenges in seamless phase II/III designs with multi-stage early stopping.
• The RB estimator is recommended for its robust performance in point estimation and confidence interval calculation.
• This approach enhances resource efficiency and statistical validity in clinical trials.